nmz0429 / biwako

The largest freshwater lake. The home of onnx models.

biwako's Issues

Add ability to look for pre-downloaded weight file on larger scope

Feature Description

It seems to be useful for user if the model can find pre-downloaded file not not only on current directory but also other places such as caches.

Plan of Implementation

1. change default path to download weight to cache
2. add another target of searching files

Resources

E.g. links, documents...

TODO

• make function to build cache dir
• download function now look for that directory as well

Add FastAPI Integration

Issue

Add FastAPI pre-defined code in api module.

Target

Sample minimum implementation of how to use this repository with FastAPI for web application.

Resources

Will be added.

TODO

• Write draft api and test it with post and get
• Test it with Streamlit frontend
• Test for all models in this repo

Add docker support

Issue

Rather than conda environment, add docker image file.

Target

What?

Resources

Any?

TODO

• [ ]
• [ ]

[i-03] Re-enable all CIs

Issue

Enable all dead CIs

Resources

Hassle

TODO

• Build
• SonarCloud
• Formatter

[i-01]Update Dependencies

Issue

Update dependencies

Resources

N/A

TODO

• Update requirements.txt
• Test Build

Issue Tracker

Current Incident Status as of 2023-12-03 04:18:13

Incident ID	Status	Summary	Severity	Issue
i-01	Done	Update Dependencies	critical	#36
i-02	Todo	Bump Up Mkdocs	enhancement	#38
i-03	In Progress	Re-enable all CIs	enhancement	#43

pie
    title Incident Progess
    "Completed" : 1
    "In Progress" : 1
    "To Do" : 1

Validate `setup.py` contents.

Issue

setup.py lacks some dependent modules.

Target

Update setup.py to latest version.

Resources

Any?

TODO

• Check dependency clause
• Check author info and license.
• Test with blank environment

Remove or Avoid old YOLO model

Issue

Old yolo implementation requires torch.

Target

Avoid the dependence of torch.

Resources

Any?

TODO

• Find the way to prevent auto import old yolo without loosing current usability of the interface.
• Otherwise delete old yolo.

[i-02] Bump Up Mkdocs

Issue

Bump up the version of Mkdocs and purge outdated extensions

Resources

N/A

TODO

• Update mkdocs
• Check extensions compatibility

Update README before publishment

Issue

README is not ready for publishment

Target

Update README

Resources

Any?

TODO

• Mock some major repository s.t PyTorch Lightning
• Add link to Docs

Add video inference wrapper

Issue

Add some higher level API for video

Target

With this, user can

1. pass mp4 video
2. Assign models
3. Obtain result rendered on the input video

Resources

Any?

TODO

• Implement from video to images loader
• Write wrapper which takes model as an argument
• Write stacked prediction on the single video
• Export it

Video Prediction API does not accept some optional args for rendering

Issue

The API calls render() without some optional arguments.

Resources

Any?

TODO

• add **kwargs to render method of all classes
• Add functionality to API for passing all opt args

Add some sample output images/video

Issue

Add sample output/video for demo purpose

What?

Probably one image per model and some videos for selected models

Resources

Any?

TODO

• Think of good examples
• Save it on demo folder
• Add link to them on docs site

Add flexibility to the render methods of human_attribute and resnet

The current implementation uses fixed positions for writing text on the image. Add user configurability for that.

TQDM is wrongly animated

Issue

Tqdm for weight download is not animated properly.

Target

The progress bar for weight download is not properly updated while download is in place.

Resources

Sample implementation.

https://blog.narito.ninja/detail/66

TODO

• Update tqdm
• Test it with various files

New Mode: NanoDet

Issue

Add lightweight object detector NanoDet

Target

Implement NanoDet onnx module

Resources

official repo

TODO

• Export model
• Upload it to release
• Write with base_interface
• Test live demo

Publish docs site

Issue

Publish Documentation by Github Pages

Target

Publish documentation which satisfies

1. Per model API reference
2. Overall explanation
3. Some images for rendering functionality

Resources

MkDocs/gh actions

TODO

• Test docs site locally
• Deploy it

Add functionality to measure runtime

Issue

Add option to measure runtime of inference

Target

For predict() and possibly render() add ability to measure runtime theirselves.
Since it is likely too exhaustive to modify all such methods, add new method or attribute to base class.

Resources

Any?

TODO

• Think of most efficient implementation
• Implement it
• Test

Add ability to download weight to any directory and look for the predownloaded weight.

Current logic only looks for the cwd and auto download it to there if not found.

Add Object Tracking

Issue

Add Object Tracking post processor for object detection module.

Target

1. Compatibility to the current YOLO implementation
2. Independence from other module.

Resources

See ~/github/onnx_byte for current progress

TODO

• Design the module with compatibility
• Implement it without pytorch
• Test it with YOLO

Add Edge Detection

Issue

Add Edge Detection Model

Target

Add some edge detection model and its visualization code.

Resources

Any?

TODO

• Write in Base_Inference
• Test with live_demo.py

Remove Dependence on PyTorch from Object Detection

Objective

Current implementation uses PyTorch for NMS. Remove this dependence by replacing corresponding lines to cv2.dnn.nms

TODO

1. Replace nms to cv2.dnn.nms
2. Fix relevant lines
3. Test for performance

Add Utility Method to Display Model Availability

Feature Description

Add some way for user to see which models (and weight types) are available.

Plan of Implementation

Something like

BiWAKO.YOLO.print_weights()

Resources

E.g. links, documents...

TODO

• Design a method as an user friendly interface
• Implement it for base inference class so that every model can use it